lkml.org 
[lkml]   [2020]   [Feb]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    From
    Subject[PATCH v2 0/4] tracing: synth_event_trace fixes
    Date
    Hi Steve,

    This is v2 of this patchset. It adds a new patch, 'tracing: Fix
    number printing bug in print_synth_event()' that fixes the problem you
    reported with (null) printing in the synth event trace output on
    32-bit systems. It turns out to be nothing to do with these fixes or
    the new stuff in 5.6 - the problem exists in 5.5 as well and can also
    cause oopses as well when reading the trace file when synthetic evens
    are present.

    The only other change from v1 is a change to 'tracing: Check that
    number of vals matches number of synth event fields' to not directly
    return -EINVAL, but only after calling __synth_event_trace_end().

    I've tested on both 32-bit x86 and x86_64, both the synth module test
    and selftests and everything looks ok here.

    Thanks,

    Tom

    v1 text:

    Sorry, it took me some time to get a 32-bit x86 system up and running
    here in order to build and test things on i386. These patches pass
    both selftests and the synth_event_gen_test testing, although the bug
    where (null) prints after every integer field in the trace output is
    still there and is there even before these or yesterday's patches - I
    have a suspicion it's been there for awhile but nobody looked at
    synthetic event trace output on i386. In any case, I'm going to
    continue looking into that - it's a weird situation where nothing gets
    put in the final %s in the format string on i386 so shows as (null),
    even though it looks like it's there. Anyway..

    Here are 3 bugfix patches, the first of which fixes the bug seen by
    the test robot, and the other two are patches that fix a couple things
    I noticed when doing the first patch.

    The previous patch I sent, changing u64 to long for the test robot bug
    did fix that problem too, but on i386 systems that would reduce every
    field to 32 bits, which isn't what we want either. The new patch
    doesn't change the code in synth_event_trace() - it still uses u64
    just like synth_event_trace_array() which takes an array of u64.
    Without any further information such as a format string, I don't know
    of a better way to deal with the varargs version, other than require
    it get passed what it expects, u64 params.

    The second patch adds the same endianness fix as for
    trace_event_raw_event_synth(), and the last one just adds back a
    missing check fot synth_event_trace() and synth_event_trace_array().

    Thanks,

    Tom

    The following changes since commit 359c92c02bfae1a6f1e8e37c298e518fd256642c:

    Merge tag 'dax-fixes-5.6-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm (2020-02-11 16:52:08 -0800)

    are available in the git repository at:

    git://git.kernel.org/pub/scm/linux/kernel/git/zanussi/linux-trace.git ftrace/synth-event-gen-fixes2-v2

    Tom Zanussi (4):
    tracing: Make sure synth_event_trace() example always uses u64
    tracing: Make synth_event trace functions endian-correct
    tracing: Check that number of vals matches number of synth event
    fields
    tracing: Fix number printing bug in print_synth_event()

    kernel/trace/synth_event_gen_test.c | 34 ++++++------
    kernel/trace/trace_events_hist.c | 108 +++++++++++++++++++++++++++++++++---
    2 files changed, 116 insertions(+), 26 deletions(-)

    --
    2.14.1

    \
     
     \ /
      Last update: 2020-02-14 23:57    [W:4.121 / U:0.052 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site